Observing the memory mills running

Last year we have explored how to make sense of where RAM is being used in Linux, including looking at all the information that /proc/meminfo provides.

Since the memory usage is rarely static, the memory management subsystem has to satisfy the memory allocation requests coming from both userspace and other kernel subsystems, while not running out of the available RAM. This results in performing various actions such as memory reclaim and compaction.

In this talk we will look at how these actions can be observed, mainly from the event counters provided in /proc/vmstat, and put those counters in the bigger context. This knowledge can help determining what might be wrong when your system is not performing well in low-memory conditions.